Antagonistic functions between the RNA chaperone Hfq and an sRNA regulate sensitivity to the antibiotic colicin

Abstract

The RNA chaperone Hfq is a key regulator of the function of small RNAs (sRNAs). Hfq has been shown to facilitate sRNAs binding to target mRNAs and to directly regulate translation through the action of sRNAs. Here, we present evidence that Hfq acts as the repressor of cirA mRNA translation in the absence of sRNA. Hfq binding to cirA prevents translation initiation, which correlates with cirA mRNA instability. In contrast, RyhB pairing to cirA mRNA promotes changes in RNA structure that displace Hfq, thereby allowing efficient translation as well as mRNA stabilization. Because CirA is a receptor for the antibiotic colicin Ia, in addition to acting as an Fur (Ferric Uptake Regulator)-regulated siderophore transporter, translational activation of cirA mRNA by RyhB promotes colicin sensitivity under conditions of iron starvation. Altogether, these results indicate that Fur and RyhB modulate an unexpected feed-forward loop mechanism related to iron physiology and colicin sensitivity.

Introduction

To successfully occupy their specific niche, bacteria have developed strategies to prevent growth of competitors. One of these strategies is the production of bacteriocins that are protein toxins produced by most bacteria to compete with similar or closely related strains (Riley and Wertz, 2002). Colicins are E. coli-specific bacteriocins released in the environment to kill other E. coli strains. They have been shown to play key roles in promoting diversity and coexistence of bacterial populations in the mammalian colon (Kerr et al, 2002; Kirkup and Riley, 2004). Colicins are divided into two subgroups according to the energy-transducing system they use to invade targeted cells. Whereas group A colicins use the Tol system, group B colicins use the Ton system (Lazdunski et al, 1998; Braun et al, 2002). Colicins exert their activity through a variety of mechanisms ranging from pore formation in the inner membrane to nuclease activity against DNA or RNA (Kleanthous, 2010).

Colicin gene clusters are generally carried on a plasmid and typically encode an immunity protein, which is expressed constitutively to protect the cell from its own colicin attack. The cluster also encodes a lysis protein that allows the release of the colicin in the environment through cell lysis (Riley, 1993a, 1993b). Once released, colicins bind with high affinity to outer membrane receptors of the target cells. Proteins recognized by colicins range from vitamin transporter to siderophore receptors (e.g., CirA, FepA, and FhuA).

One of the most extensively studied pore-forming colicin is colicin Ia, which targets the outer membrane protein (OMP) CirA (Buchanan et al, 2007; Jakes and Finkelstein, 2010). Colicin Ia is a 69-kDa protein that inserts into the inner membrane of the target cell to form a channel responsible for cell death (Wiener et al, 1997). CirA is a TonB-dependent transporter involved in the uptake of ferric iron (Fe³⁺) complexed with catechol siderophores such as dihydroxybenzoate (DHB) and dihydroxybenzoyl serine (DHBS), which are respectively precursor and breakdown product of the siderophore enterobactin (Hantke, 1990). As for most genes involved in iron uptake, cirA transcription is repressed by the Fur (Ferric Uptake Regulator) protein bound to ferrous iron (Fe²⁺) (Griggs et al, 1987). Moreover, two redundant small RNAs (sRNAs), OmrA and OmrB, have also been shown to repress cirA translation under conditions of high osmolarity (Guillier and Gottesman, 2006, 2008).

Bacterial sRNAs are key regulators of cellular functions by modulating gene expression in response to various environmental cues (Gottesman and Storz, 2010). These sRNAs regulate target mRNAs by direct base pairing to positively or negatively affect their translation and stability (Storz et al, 2011). In most cases, sRNAs require the RNA chaperone Hfq for optimal regulation by promoting sRNA-mRNA pairing and by stabilizing some sRNAs in vivo (Vogel and Luisi, 2011; Andrade et al, 2012). Hfq can also act as a translational repressor by competing directly with initiating 30S ribosomal subunit for accessibility to the ribosome binding site (RBS) (Vytvytska et al, 2000; Desnoyers and Massé, 2012).

The sRNA RyhB, which regulates iron homeostasis, is one of the most studied Hfq-dependent sRNAs. Under iron-rich conditions, Fe²⁺-Fur represses ryhB transcription. In contrast, during iron starvation conditions, Fur becomes inactive and relieves repression of ryhB (Massé and Gottesman, 2002; Salvail and Massé, 2011). Under these conditions, RyhB directly regulates ∼20 different mRNAs encoding iron-using proteins. By binding to those mRNAs, RyhB shuts down translation and stimulates their rapid degradation through the action of RNase E (Massé et al, 2003; Massé et al, 2005). RyhB also promotes siderophore production through repression of cysE mRNA that encodes a serine acetyltransferase, which results in an increased flux of serine into enterobactin production (Salvail et al, 2010).

In addition, RyhB can act as a gene activator. Indeed, RyhB activates translation of the Fur-independent shiA mRNA that encodes a transporter of shikimate, an intermediate in the synthesis of enterobactin (Prévost et al, 2007). Hfq binds the 5′-untranslated region (UTR) of shiA and potentially promotes the formation of an inhibitory structure that sequesters the translation initiation region (TIR) and limits translation. However, under conditions of iron starvation, RyhB base pairs with shiA mRNA to disrupt the inhibitory structure, thereby favouring translation and transcript stabilization. This mechanism is reminiscent of similar cases of translational activation by sRNAs such as activation of rpoS by DsrA (Sledjeski et al, 1996; McCullen et al, 2010) and activation of glmS mRNA by GlmZ sRNA (Urban and Vogel, 2008).

In this work, we present evidence that RyhB is a novel regulator of cirA expression. We showed that RyhB expression was essential for CirA synthesis during iron starvation. Our data further suggested that in the absence of RyhB, Hfq repressed cirA mRNA translation, thereby causing rapid transcript turnover through the action of RNase E and low accumulation of CirA. RyhB pairing to cirA mRNA activated its translation and prevented its destabilization by RNase E. The resulting increased levels of CirA transporter made the cells sensitive to the bactericidal action of colicin Ia. This new RyhB-mediated regulation was unexpected considering that cirA is, to our knowledge, the first Fur-regulated gene to require post-transcriptional activation by RyhB to be expressed during iron starvation. The Fur-RyhB-cirA regulatory circuit forms a coherent feed-forward loop (Fur represses cirA and ryhB while RyhB activates cirA), predicted to confer altered regulatory dynamics in comparison to direct regulation by Fur (Mangan and Alon, 2003; Beisel and Storz, 2010). In addition to its role as a gene silencer, our results confirmed the role of RyhB in gene activation.

Results

RyhB expression is essential for CirA production

Previous microarray results have shown that RyhB expression resulted in a 3.3-fold increase in cirA transcript levels in a Δfur ΔryhB background (Massé et al, 2005). To extend these results, we used quantitative real-time PCR (qRT–PCR) to monitor cirA mRNA levels after a 20-min expression of RyhB in ΔryhB and Δfur ΔryhB backgrounds, as in the case of the microarray study. Results showed that cirA transcript levels increased 4.7-fold upon RyhB expression in the Δfur ΔryhB background (Figure 1A). This increase was higher than the previously characterized RyhB-positive target shiA (2.3-fold increase) (Prévost et al, 2007). Intriguingly, RyhB expression in a ΔryhB background resulted in a 3.3-fold decrease in cirA transcript levels. We have previously reported that RyhB expression under these conditions promoted an increase in free intracellular iron concentration (Jacques et al, 2006). Thus, the decrease in cirA mRNA levels observed here may result from Fe²⁺-Fur repression, as observed in the case of Fur-regulated genes (Massé et al, 2005). In contrast, RyhB still upregulated the expression of the Fur-independent shiA gene in the ΔryhB background (Figure 1A).

Figure 1.

RyhB expression is essential for CirA production. (A) Quantitative RT-PCR (qRT-PCR) analysis of RyhB effect on cirA and shiA mRNA levels. Strains EM1455 (ΔryhB) and EM1493 (Δfur ΔryhB), each carrying either pBAD-ryhB or control plasmid pNM12, were grown in LB medium. Arabinose was added (0.1%, total concentration) to an OD₆₀₀ of 0.5, followed by total RNA extracted after 20 min. Primers complementary to cirA and shiA open reading frames (ORFs) were used for qRT–PCR assays. Mean and standard deviation (s.d.) values of triplicate samples are shown. (B) Northern blot analysis of RyhB effect on cirA mRNA levels. Strain EM1493 (Δfur ΔryhB) carrying pBAD-ryhB or control plasmid pNM12 was grown in LB medium. Arabinose was added (0.1%, total concentration) to an OD₆₀₀ of 0.5 and total RNA was extracted at the indicated time points. A probe complementary to cirA ORF was used. (C) Northern blot analysis of RyhB and Fur effects on cirA mRNA levels in cells grown under iron-poor conditions. Strains EM1055 (WT), EM1238 (ΔryhB), KP392 (Δfur) and KP393 (Δfur ΔryhB) were grown in M63 iron-free glucose minimal medium and total RNA was extracted at the indicated values of OD₆₀₀. A probe complementary to cirA ORF was used. (D) Primer extension analysis of RyhB effect on cirA mRNA stability. Strain EM1493 (Δfur ΔryhB) carrying pBAD-ryhB or control plasmid pNM12 was grown in LB medium and arabinose was added (0.1%, total concentration) to an OD₆₀₀ of 0.5 (−10 time point). Rifampicin (250 μg/ml, final concentration) was added 10 min after the addition of arabinose (0 time point) and total RNA was extracted at the indicated time points. Primers complementary to cirA (EM1408) and 16S rRNA (EM345) were used to perform primer extension reactions on total RNA samples. Primer extension signals of two independent experiments were quantified by densitometry and normalized to time zero (0) for both pNM12 and pBAD-ryhB. Mean and s.d. values of duplicate experiments are shown. (E) Coomassie-stained SDS gel of outer membrane proteins (OMPs) extracts from KP392 (Δfur), HS437 (Δfur ΔcirA), KP393 (Δfur ΔryhB) and MPC253 (Δfur ΔryhB Δhfq) cells grown in LB medium at an OD₆₀₀ of 1.0. Molecular sizes of co-migrating proteins (L) are indicated at the left in kDa.

Source data for this figure is available on the online supplementary information page.

Source Data for Figure 1 ^{(654.6KB, pdf)}

Full-length cirA mRNA expression was monitored by northern blot following pulse expression of RyhB in the Δfur ΔryhB background. This showed that cirA transcript accumulation following RyhB expression was swift with maximal transcript accumulation following 8 min of sRNA induction (Figure 1B, lane 5). However, despite high cirA promoter activity in the Δfur ΔryhB background (Supplementary Figure S1A, see pNM12), cirA transcript poorly accumulated in the absence of RyhB (Figure 1B, lanes 8–14). Moreover, RyhB had no significant effect on cirA promoter (Supplementary Figure S1A, pNM12 and pBAD-ryhB), suggesting that RyhB regulated cirA post-transcriptionally.

Since RyhB is naturally expressed during iron starvation, we next studied the endogeneous effect of RyhB on cirA expression in iron-free minimal medium. Results showed that cirA mRNA levels in WT cells correlated with RyhB levels (Figure 1C, lanes 1 and 2). In contrast, ΔryhB cells did not accumulate cirA transcripts at any time (Figure 1C, lanes 3 and 4), despite promoter activity similar to WT cells (Supplementary Figure S1B, WT and ΔryhB). Inactivation of fur resulted in increased levels of cirA mRNA as compared to WT cells (Figure 1C, lanes 1 and 2, and lanes 5 and 6). This behaviour probably resulted from increased cirA promoter activity (Supplementary Figure S1B, WT and Δfur) due to the absence of Fe²⁺-Fur repression. The absence of Fur repression also resulted in higher levels of RyhB in Δfur cells as compared to WT cells (Figure 1C, lanes 1 and 2, and lanes 5 and 6) which may also contribute to increased levels of cirA mRNA. Strikingly, inactivation of ryhB in a Δfur background resulted in a severe decrease in cirA transcript levels (Figure 1C, compare lanes 5 and 6 with lanes 7 and 8), despite an increase in cirA promoter activity (Supplementary Figure S1B, Δfur and Δfur ΔryhB, OD₆₀₀ of 0.6). Taken together, these results suggested that RyhB expression was essential to promote cirA transcript accumulation upon iron starvation.

Because our data indicated that RyhB was essential for cirA expression, we hypothesized that RyhB could stabilize cirA mRNA. To address this, we determined the half-life of cirA mRNA under conditions where RyhB was expressed (pBAD-ryhB) or not (pNM12). Because full-length cirA accumulated poorly in the Δfur ΔryhB background in LB medium (Figure 1B), we performed a primer extension using a reverse primer complementary to the beginning of cirA coding sequence. This approach was found to be sensitive enough to observe cirA mRNA expression in the absence of RyhB (Figure 1D, lane 2). While significant amounts of cirA transcripts were detected after 8 min in the presence of RyhB (Figure 1D, lane 12), most of the cirA mRNA disappeared after 2 min in the absence of RyhB (Figure 1D, lane 3). Densitometry analysis revealed that cirA mRNA had a half-life of 5 min when RyhB was expressed (pBAD-ryhB) but of less than 2 min in the absence of RyhB (Figure 1D).

Since RyhB promoted cirA mRNA accumulation, we next sought to determine the effect on CirA protein expression. Because CirA is an OMP, we compared the OMP profiles of Δfur, Δfur ΔcirA, Δfur ΔryhB, and Δfur ΔryhB Δhfq cells grown in LB media. Comparison of the profiles between Δfur and Δfur ΔryhB cells revealed the presence of a protein of ∼65 kDa for which the levels of expression were very low in Δfur ΔryhB cells as compared to Δfur cells (Figure 1E, lanes 1 and 3). Intriguingly, the absence of Hfq in Δfur ΔryhB Δhfq cells appears to improve the expression of the 65-kDa protein as compared to Δfur ΔryhB cells (Figure 1E, compare lanes 3 and 4). Because this protein was absent in Δfur ΔcirA cells (Figure 1E, lane 2) and since the estimated molecular weight of CirA is between 54 and 74 kDa (Buchanan et al, 2007), we assumed this protein to be CirA. The bulk of these results suggested that RyhB was essential for CirA production.

RyhB directly activates cirA in vivo

Data reported above (Figure 1) suggested that RyhB activated cirA through direct pairing with its mRNA. Bioinformatic analysis using the TargetRNA software (Tjaden et al, 2006) suggested potential pairing of RyhB with cirA 5′-UTR in the region −58 to −41 relative to the AUG start codon (Figures 2 and 3A). We then proceeded to investigate this potential RyhB pairing site by chemical in vitro analysis. A 5′-end-labelled cirA transcript containing the whole 5′-UTR and 79 additional nucleotides (nt) from the coding region was incubated in the presence or absence of a 20-fold molar excess of unlabelled RyhB, followed by a partial digestion by lead acetate (PbAc, cleaves unpaired residues). Results showed that the addition of RyhB (R) protected from cleavage (Figure 3B, lanes 6 and 7) in the region −42 to −51 that corresponded to the predicted RyhB pairing core (Figures 2 and 3A).

Figure 2.

The 5′-UTR of cirA mRNA potentially forms an inhibitory structure preventing translation initiation. RNA-fold prediction of cirA 5′-UTR secondary structure (nucleotides −173 to +18 relative to the AUG start codon). The RyhB pairing core, the Hfq binding sites, the M2G mutation of Hfq binding site II and the MHfqIII mutation of Hfq binding site III are indicated. Probing results presented in Figure 7B and Supplementary Figure S7 are summarized here. Stem-loop structures are numbered (SL1, SL2, and SL3).

Figure 3.

RyhB base pairs with cirA mRNA in vivo. (A) TargetRNA prediction of pairing between RyhB and cirA mRNA. The cirA1 and RyhB1 mutations used in (B–D) are indicated. (B) Lead acetate (PbAc) probing of RyhB pairing with cirA mRNA. 5′-end-labelled cirA or cirA1 mRNAs (nucleotides −173 to +78 relative to the AUG start codon) were incubated in the absence or in the presence of RyhB (R) or RyhB1 (R1) before addition of PbAc. C, non-reacted controls; OH, alkaline ladder; T1, RNase T1 ladder (guanine residues). (C) Northern blot analysis of arabinose-induced RyhB1 effect on cirA mRNA levels. Strain EM1493 (Δfur ΔryhB) carrying pBAD-ryhB, pBAD-ryhB1 or control plasmid pNM12 was grown in LB medium. Arabinose (0.1%, total concentration) was added to an OD₆₀₀ of 0.5 and total RNA was extracted at the indicated time points. Probes complementary to cirA and sodB open reading frames (ORFs) were used. (D) Effect of arabinose-induced RyhB and RyhB1 on cirA′-′lacZ and cirA1′-′lacZ translational fusions in a Δfur ΔryhB background. Cells were grown in LB medium and arabinose (0.1%, total concentration) was added to an OD₆₀₀ of 0.1. Specific β-galactosidase activity from three independent cultures was then measured 3 h later. Mean and standard deviation (s.d.) values are shown. Empty vector pNM12 was used as a control.

Source data for this figure is available on the online supplementary information page.

Source Data for Figure 3 ^{(1.5MB, pdf)}

To further confirm that RyhB paired directly with cirA mRNA in vivo, we constructed the RyhB1 allele in which nucleotide C47 was exchanged for G47 to decrease the strength of pairing with cirA (see details in Figure 3A). The cleavage assay showed that the addition of RyhB1 (R1) did not protect bases that were determined to be part of the RyhB pairing core (Figure 3B, lanes 6 and 8). These data strongly suggest that RyhB1 was not able to base pair with cirA mRNA in vitro. RyhB1 was then expressed using an arabinose-inducible vector and its effect on cirA mRNA levels was monitored by northern blot analysis. Results showed that in spite of similar levels of accumulation compared to RyhB (Figure 3C, lanes 5 and 6, and lanes 8 and 9), RyhB1 failed to promote cirA expression within 15 min as opposed to RyhB expression that resulted in rapid accumulation of cirA transcripts (Figure 3C, lanes 4–6, and lanes 7–9). However, RyhB1 was still able to repress sodB, an RyhB-repressed gene (Figure 3C, lanes 7–9). This observation suggested that RyhB1 remained functional and that the loss of cirA regulation probably reflected a disruption of target-specific base pairing.

To further characterize RyhB-cirA interaction, we constructed the cirA1 allele in which nucleotide G47 was changed to C47 to restore base pairing with RyhB1 (see Figure 3A for details). Results indicated that RyhB addition did not promote cleavage protection of cirA1 (Figure 3B, lanes 14 and 15), indicating that RyhB was not able to base pair with cirA1 mRNA. As predicted, the compensatory mutation introduced in RyhB1 was able to restore base pairing with cirA1 since the nucleotides of cirA1 that are part of the RyhB pairing core were protected from cleavage (Figure 3B, lane 14 and lane 16). We next performed in vivo covariation mutagenesis using cirA′-′lacZ and cirA1′-′lacZ translational fusions, harbouring region −280 to +33 with respect to the initiation codon. Results indicated that RyhB expression resulted in a four-fold activation of the cirA′-′lacZ translational construct but failed to activate cirA1′-′lacZ fusion while RyhB1 only activated the complementary cirA1′-′lacZ construct without affecting the cirA′-′lacZ construct (Figure 3D). Taken together, these data confirmed that RyhB promoted CirA synthesis through direct pairing with cirA mRNA.

RyhB protects cirA mRNA from RNase E degradation

Since RNase E is known to be the main endonuclease responsible for RNA turnover in E. coli (Belasco, 2010), we explored the possibility that RyhB protected cirA mRNA from RNase E degradation. Two different mutants of RNase E were used and their effect on cirA transcript levels in the presence or absence of RyhB expression was investigated. One mutant (rne131) lacked residues 586–1061 that constitute the C-terminal domain involved in interaction with other components of the RNA degradosome and with Hfq (Kido et al, 1996; Vanzo et al, 1998). Results showed that cirA mRNA levels were low in ΔryhB as in rne131 ΔryhB cells (Supplementary Figure S2A, lanes 2 and 4). These data led us to conclude that the degradation of cirA mRNA in the absence of RyhB did not require interaction of RNase E with the RNA degradosome components or with Hfq. The other mutant (rne-3071) was a thermosensitive allele where RNase E becomes inactive at non-permissive temperature (44°C for 15 min) (McDowall et al, 1993). In these experiments, cirA transcript accumulated at levels similar to WT in the absence of RyhB expression when RNase E was inactivated following a 15-min exposure of the rne-3071 ΔryhB strain at 44°C (Supplementary Figure S2B, lanes 6 and 8). These data indicate that RNase E was responsible for the turnover of cirA mRNA in the absence of RyhB.

RyhB stabilizes cirA mRNA through translational activation

RNase E is considered to be responsible for the rapid decay of translationally impaired mRNAs (Arnold et al, 1998; Deana and Belasco, 2005). We thus tested the possibility that RyhB protected cirA mRNA from RNase E degradation by activating its translation, as reported in the case of shiA mRNA (Prévost et al, 2007). This possibility was investigated by first comparing the effect of RyhB on the activity of cirA′-lacZ transcriptional fusion and cirA′-′lacZ translational fusion. Results demonstrated that RyhB expression resulted in a 1.6-fold activation of transcriptional cirA′-lacZ fusion and a 3.8-fold activation of translational cirA′-′lacZ fusion (Figure 4A). Moreover, we compared the activities of both fusions in WT and ΔryhB cells grown in an iron-free minimal medium. The absence of ryhB resulted in a 2.2-fold decrease in activity for transcriptional cirA′-lacZ fusion and a 4.1-fold decrease in activity for translational cirA′-′lacZ fusion (Figure 4B). The differential effect of RyhB expression on translational fusion as compared to transcriptional fusion suggested that RyhB activated cirA translation.

Figure 4.

RyhB protects cirA mRNA from degradation by activating its translation. (A) Effect of arabinose-induced RyhB on cirA′-lacZ transcriptional fusion and on cirA′-′lacZ translational fusion in a Δfur ΔryhB background. Cells were grown in LB medium and 0.1% arabinose was added at an OD₆₀₀ of 0.1. Specific β-galactosidase activity from three independent cultures was then measured 3 h later. Mean and standard deviation (s.d.) values are shown. Empty vector pNM12 was used as a control. (B) β-Galactosidase assay of cirA′-lacZ transcriptional fusion and cirA′-′lacZ translational fusion in WT and ΔryhB cells grown in M63 iron-free glucose minimal medium at an OD₆₀₀ of 0.6. Specific β-galactosidase activity from three independent cultures was measured. Mean and s.d. values are shown. (C) Effect of arabinose-induced RyhB on cirAmutAUG′-lacZ transcriptional fusion in a Δfur ΔryhB background. Cells were grown in LB medium and 0.1% arabinose was added at an OD₆₀₀ of 0.1. Specific β-galactosidase activity from three independent cultures was then measured 3 h later. Mean and s.d. values are shown. Empty vector pNM12 was used as a control. (D) β-galactosidase assay of cirAmutAUG′-lacZ transcriptional fusion in WT and ΔryhB cells grown in M63 iron-free glucose minimal medium at an OD₆₀₀ of 0.6. Specific β-galactosidase activity from three independent cultures was measured. Mean and s.d. values are shown.

We next constructed the cirAmutAUG′-lacZ transcriptional fusion in which the cirA initiation codon AUG was changed to CUG to prevent translation initiation. As expected, β-galactosidase activity was not detected in cirAmutAUG′-′lacZ translational fusion (data not shown). RyhB expression did not activate cirAmutAUG′-lacZ transcriptional fusion as opposed to a 1.5-fold activation observed in the case of cirA′-lacZ fusion (Figure 4C). Furthermore, when cells were grown in minimal iron-free medium, there was no effect on cirAmutAUG′-lacZ transcriptional fusion whether RyhB was present or not (WT/ΔryhB=0.9) as opposed to a 2.1-fold increase in cirA′-lacZ activity observed in WT cells as compared to ΔryhB cells (Figure 4D). cirAmutAUG′-lacZ transcriptional fusion activity was six-fold lower than cirA′-lacZ fusion. We attributed the decrease in β-galactosidase activity to the instability of untranslated cirAmutAUG mRNA. Our results suggested that RyhB pairing with cirA mRNA increased translation, protected the ORF from RNase E degradation and therefore resulted in accumulation of full-length cirA mRNA.

Hfq represses cirA translation

Previously, we have reported that Hfq repressed shiA mRNA translation when RyhB was not expressed (Prévost et al, 2007). We thus investigated whether Hfq played a similar role in cirA activation by RyhB. Northern blots were used to monitor cirA transcript levels in Δhfq and Δhfq ΔryhB cells. Results revealed that hfq inactivation in a ΔryhB background fully restored cirA expression (Figure 5A, lanes 2 and 4), as observed in the case of shiA mRNA. In agreement with previous reports (Wassarman et al, 2001; Holmqvist et al, 2010), we observed that Hfq had an effect on OmrA/B expression as revealed by the fact that both sRNAs were barely detectable in Δhfq and Δhfq ΔryhB cells in comparison with WT and ΔryhB cells (Figure 5A, lanes 1 and 2, and lanes 3 and 4). With respect to this observation, the WT levels of cirA mRNA in Δhfq ΔryhB cells could be attributed to decreased OmrA/B repression. However, this is unlikely because omrAB inactivation in a ΔryhB background failed to increase levels of cirA mRNA (Supplementary Figure S3, lanes 2 and 4).

Figure 5.

Hfq represses cirA mRNA translation. (A) Northern blot analysis of Hfq effect on cirA mRNA levels. Strains EM1055 (WT), EM1238 (ΔryhB), EM1265 (Δhfq) and KP111 (Δhfq ΔryhB) were grown in M63 iron-free glucose minimal medium at an OD₆₀₀ of 0.6 and total RNA was extracted. Probes complementary to cirA and shiA open reading frames (ORFs) were used. (B) β-Galactosidase assay of cirA′-lacZ transcriptional fusion and cirA′-′lacZ translational fusion in Δhfq and Δhfq ΔryhB cells grown in M63 iron-free glucose minimal medium at an OD₆₀₀ of 0.6. Specific β-galactosidase activity from three independent cultures was measured. Mean and standard deviation (s.d.) values are shown. (C) Northern blot analysis of arabinose-induced RyhB effect on cirA mRNA levels in Δfur ΔryhB Δhfq cells. Strains HS506 (Δfur ΔryhB) and HS518 (Δfur ΔryhB Δhfq), each carrying pBAD-ryhB or control plasmid pNM12, were grown in M63 iron-free glycerol minimal medium. Arabinose (0.1%) was added at an OD₆₀₀ of 0.3 and total RNA was extracted 10 min later. A probe complementary to cirA ORF was used. (D) Effect of arabinose-induced RyhB on cirA′-lacZ transcriptional fusion and cirA′-′lacZ translational fusion in a Δfur ΔryhB Δhfq background. Cells were grown in M63 iron-free glycerol minimal medium and 0.1% arabinose was added at an OD₆₀₀ of 0.1. Specific β-galactosidase activity from three independent cultures was then measured 3 h later. Mean and s.d. values are shown. Empty vector pNM12 was used as a control.

Source data for this figure is available on the online supplementary information page.

Source Data for Figure 5 ^{(430.1KB, pdf)}

We next monitored the activity of transcriptional cirA′-lacZ and translational cirA′-′lacZ fusions in Δhfq ΔryhB cells in comparison to ΔryhB cells. Results showed that inactivation of hfq in ΔryhB cells restored WT levels of activity in both cases, suggesting that RyhB activated cirA mRNA translation by counteracting Hfq repression (Figure 5B). The greater effect of hfq inactivation on cirA′-′lacZ translational fusion (3.5-fold increase) as opposed to cirA′-lacZ transcriptional fusion (2.2-fold increase) suggested that cirA repression by Hfq occurred at the translational level.

The possibility that RyhB activated cirA solely by antagonizing Hfq repression would be incompatible with its capacity to further activate cirA in the absence of Hfq. However, pulse expression of RyhB in the Δfur ΔryhB Δhfq background resulted in a 1.8-fold increase in cirA mRNA levels within 10 min, indicating that RyhB was able to activate cirA in the absence of Hfq (Figure 5C, lanes 5 and 6, and lanes 7 and 8). To determine whether this increase in cirA mRNA resulted from translational activation, we monitored RyhB effect on cirA′-lacZ transcriptional and cirA′-′lacZ translational fusions. Results demonstrated that inactivation of hfq in the Δfur ΔryhB background resulted in a 2.1-fold increase in cirA′-lacZ transcriptional activity and in a 4.8-fold increase in cirA′-′lacZ translational activity (Figure 5D). These results were similar to the effect of RyhB on both fusions in the Δfur ΔryhB background (2-fold activation for cirA′-lacZ transcriptional fusion and 4.6-fold activation for cirA′-′lacZ translational fusion), in agreement with previous results (Figure 5B). These data confirmed that RyhB expression counteracted the repressive effect of Hfq on cirA mRNA. However, RyhB expression in Δfur ΔryhB Δhfq cells did not result in a significant increase in cirA′-lacZ transcriptional activity but a small but significant 1.1-fold increase in cirA′-′lacZ translational activity (Figure 5D). These results contrasted with those of northern blot experiments (Figure 5C) where a 1.8-fold increase in cirA transcript levels was observed following RyhB expression. Together, these results suggested that Hfq destabilized cirA mRNA through translational repression and that RyhB expression abolished this repression.

Hfq inhibits 30S ribosomal subunit binding to cirA mRNA upon translation initiation

We next sought to decipher the mechanism by which Hfq repressed cirA mRNA translation. Previous studies had unveiled the capacity of Hfq to inhibit 30S translation initiation complex formation on some mRNAs through binding to AU-rich single-stranded regions located in the vicinity of the TIR and acting as translational enhancers (Vytvytska et al, 2000; Desnoyers and Massé, 2012). To determine whether Hfq could repress cirA translation through this mechanism, we first performed toeprinting experiments to study the effect of Hfq on 30S binding to cirA mRNA. We used the Hfq-independent lpp transcript (Vytvytska et al, 1998, 2000) as a control (Supplementary Figure S4A and B). Results indicated that addition of purified 30S ribosomal subunits and tRNA^fmet to cirA mRNA (0.2 μM) resulted in a strong block of reverse transcription at positions +15/+16 downstream of the AUG start codon (Figure 6A). These observations were consistent with the formation of a translation initiation complex on cirA mRNA. Addition of increasing amounts of purified Hfq (200 nM) resulted in a 2.2-fold inhibition of initiation complex formation (Hfq:cirA molar ratio, 1:1) and in near complete inhibition at high concentration (500 nM) of Hfq (Hfq:cirA molar ratio, 2.5:1). These results suggested that Hfq binding to cirA mRNA may be sufficient to prevent translation initiation in vivo. As expected, higher concentrations of Hfq were required to prevent ribosome binding to lpp mRNA (Supplementary Figure S4A).

Figure 6.

Hfq represses 30S ribosomal subunit binding to cirA mRNA upon translation initiation. (A) Toeprint analysis of Hfq effect on 30S ribosomal subunit binding to cirA mRNA (200 nM) upon translation initiation. Unlabelled cirA mRNA was incubated in the presence of the indicated amounts of purified Hfq protein. Positions +15/+16 relative to cirA AUG start codon are indicated. CTAG refers to sequencing ladders. Toeprint signals were quantified by densitometry and data are reported in arbitrary units. (B) RNase I probing of Hfq binding sites on cirA mRNA. 5′-end-labelled cirA mRNA (nucleotides −173 to +78 relative to the AUG start codon) was incubated in the presence of the indicated amounts of purified Hfq protein before addition of RNase I. Hfq I, Hfq II, and Hfq III refer to Hfq binding sites I, II, and III (see Figure 2). C, non-reacted controls; OH, alkaline ladder; T1, RNase T1 ladder (guanine residues); TA, RNase TA ladder (adenine residues). (C) PbAc probing of Hfq binding sites on cirA mRNA. 5′-end-labelled cirA mRNA (nucleotides −173 to +78 relative to the AUG start codon) was incubated in the presence of the indicated amounts of purified Hfq protein before addition of PbAc (5 mM). Hfq II and Hfq III refer to Hfq binding sites II and III (see Figure 2). Some guanine and adenine positions are given for clarity. C, non-reacted controls; OH, alkaline ladder; T1, RNase T1 ladder (guanine residues). The region corresponding to Hfq binding site II (Hfq II) is enlarged to illustrate protection from PbAc cleavage of nucleotides −33 to −31 following addition of Hfq. (D) Toeprint analysis of 30S ribosomal subunit binding to cirAM2G mRNA. Unlabelled cirA and cirAM2G mRNAs (0.2 μM final) were incubated in the presence of 30S ribosomal unit and tRNA^fmet. Positions +15/+16 relative to cirA and cirAM2G AUG start codons are indicated. CTAG refers to sequencing ladders. Toeprint signals of two independent experiments were quantified by densitometry and normalized to WT toeprint signal (lane 6). Data are reported in arbitrary units. Mean and standard deviation (s.d.) values of duplicate experiments are shown.

Source data for this figure is available on the online supplementary information page.

Source Data for Figure 6 ^{(2.7MB, pdf)}

We next performed enzymatic and chemical probing using RNase I (cleaves unpaired nucleotides) and PbAc to identify potential Hfq binding sites in the 5′-UTR of cirA mRNA. Addition of increasing amounts of Hfq to 5′-end-labelled cirA mRNA resulted in protection of nucleotides −105 to −91 from RNase I cleavage (Figure 6B, Hfq I site, lane 6 and lanes 7–9). This region is AU-rich and single-stranded, according to its susceptibility to RNase I cleavage (Figure 6B, lane 6, Hfq I site). These properties were consistent with previous descriptions of Hfq binding sites (Lorenz et al, 2010; Balbontín et al, 2010). Protection of nucleotides −63 to −53 from RNase I cleavage was also observed at high concentrations of Hfq (Figure 6B, lanes 6 and lanes 7–9) that may result from Hfq binding to site I.

Interestingly, incubation of cirA with Hfq resulted in a weak but noticeable protection of nucleotides +4 to +6 from RNase I cleavage (Figure 6B, Hfq III site, lanes 6–9). These nucleotides are also protected from PbAc cleavage in the presence of Hfq (Figure 6C, Hfq III site, lanes 5–8). This region is U-rich and single-stranded according to secondary structure prediction (Figure 2) and to the susceptibility of nucleotides +4 to +6 to RNase I (Figure 6B, Hfq III site, lane 6) and PbAc (Figure 6C, Hfq III site, lane 5) cleavages in the absence of Hfq. Binding of this U stretch by Hfq could potentially hamper 30S binding to the AUG start codon of cirA mRNA upon translation initiation (Figure 6A).

Nucleotides −33 to −31, which are part of an AU-rich region, were also protected from PbAc cleavage upon addition of Hfq (Figure 6C, Hfq II site, lanes 5–8). These nucleotides were predicted to be part of a stem-loop structure (Figure 2, SL3, nucleotides A-40 to G+12), consistent with their protection from RNase I cleavage (Figure 6B, Hfq II site, lane 6). These nucleotides (−33 to −31) are also part of an AU-rich region (nucleotides A-35 to U-26) located a few nucleotides upstream of the Shine-Dalgarno sequence, suggesting that this region acts as a translational enhancer. To investigate this, we constructed a cirAM2G mRNA in which nucleotides U-32 and U-31 were replaced by two guanine residues (Figure 2) and 30S binding was monitored by toeprinting. Results showed that there was ∼30% less binding of 30S to cirAM2G mRNA than to cirA mRNA (Figure 6D, lanes 6 and 7), suggesting an important role of region A-35 to U-26 in translation initiation.

Despite the fact that Hfq I site was highly AU-rich (Figure 2), we did not expect this region to act as a translational enhancer. The Hfq I site is too far upstream of the Shine-Dalgarno sequence in comparison with translational enhancer elements previously described (Hook-Barnard et al, 2007; Desnoyers and Massé, 2012). Indeed, replacement of nucleotides A-100 and A-98 for cytosine residues (cirAML, Supplementary Figure S5A) did not result in reduced formation of 30S translation initiation complex (Supplementary Figure S5B, lanes 7 and 8). Taken together, our results suggested that Hfq may repress cirA mRNA translation by preventing 30S initiation complex formation through binding in the vicinity of the AUG start codon (Hfq III site) and/or to a translational enhancer element (Hfq II site) located a few nucleotides upstream of the Shine-Dalgarno sequence.

RyhB promotes 30S ribosomal subunit binding to cirA mRNA in the presence of Hfq

Data presented above (Figure 4) strongly suggested that RyhB stabilized cirA mRNA through translational activation. We hypothesized that RyhB pairing with cirA 5′-UTR would result in increased 30S binding to the AUG start codon upon translation initiation. Unexpectedly, addition of increasing amounts of RyhB to cirA mRNA (0.2 μM) resulted in a 12% inhibition of initiation complex formation in the presence of RyhB (0.4 μM, RyhB:cirA molar ratio, 2:1) and a 30% inhibition with a higher concentration (0.8 μM) of RyhB (RyhB:cirA molar ratio, 4:1) (Supplementary Figure S6, lane 6 and lanes 8–10). This negative effect of RyhB on 30S binding to cirA mRNA appeared to be pairing specific since addition of RyhB1 did not repress initiation complex formation (Supplementary Figure S6, lane 6 and lanes 12 and 14).

As a follow-up to results presented above (Figure 5B) that suggested that RyhB activated cirA by counteracting Hfq translational repression, we tested the effect of RyhB on 30S binding to cirA mRNA in the presence of Hfq. Addition of Hfq (150 nM, Hfq:cirA molar ratio, 0.75:1) resulted in 30% inhibition of 30S binding to cirA mRNA (Figure 7A, lanes 6 and 8). Consistent with the in vivo data, addition of RyhB (150 nM) in the presence of Hfq (Hfq:RyhB:cirA molar ratio, 0.75:0.75:1) resulted in a 60% increase in the toeprint signal (Figure 7A, lanes 8 and 10), whereas addition of the same amount of RyhB1 only results in a 25% increase in 30S binding (Figure 7A, lanes 8 and 12). The weak RyhB1-dependent activation may be attributable to Hfq displacement from cirA mRNA via RyhB1 poly(U) tail, which has previously been shown to bind Hfq with high affinity (Otaka et al, 2011). We could not exclude the possibility that this titration effect may contribute to RyhB-positive effect on 30S binding to cirA mRNA (Figure 7A, lanes 8 and 10). However, the fact that RyhB induced stronger toeprint signals than those observed upon addition of RyhB1 (Figure 7A, lanes 10 and 12), strongly suggested that the RyhB-dependent activation was base pairing specific. Moreover, RyhB did not affect ribosome binding to lpp mRNA (negative control) in the presence of Hfq (Supplementary Figure S4B, compare lanes 8 and 10). Taken together, toeprinting data supported the interpretation that RyhB directly promoted translational initiation complex formation on cirA mRNA in the presence of Hfq.

Figure 7.

RyhB promotes 30S ribosomal subunit binding to cirA mRNA in the presence of Hfq. (A) Toeprint analysis of RyhB and RyhB1 effects on 30S ribosomal subunit binding to cirA mRNA in the presence of Hfq upon translation initiation. Unlabelled cirA mRNA (0.2 μM final) was incubated in the presence of Hfq (0.15 μM) and RyhB (0.15 μM) or RyhB1 (0.15 μM). Positions +15/+16 relative to cirA AUG start codon are indicated. CTAG refers to sequencing ladders. Toeprint signals of two independent experiments were quantified by densitometry and normalized to the toeprint signal in lane 6. Data are reported in arbitrary units. Mean and standard deviation (s.d.) values of duplicate experiments are shown. (B) In-line probing analysis of Hfq and RyhB effects on cirA mRNA. 5′-end-labelled cirA mRNA (nucleotides −173 to +78 relative to the AUG start codon) was incubated for 40 h in the presence of Hfq (0.3 μM) and RyhB (1 μM). Some guanine positions are illustrated for clarity. C, non-reacted controls; OH, alkaline ladder; T1, RNase T1 ladder (guanine residues).

Source data for this figure is available on the online supplementary information page.

Source Data for Figure 7 ^{(1.5MB, pdf)}

We next investigated the mechanism by which RyhB stimulated ribosome binding to cirA mRNA in the presence of Hfq upon translation initiation. Previous studies of sRNA-mediated translational activation (e.g., RyhB-shiA, DsrA-rpoS, and GlmZ-glmS) have shown that sRNA pairing with its target mRNA usually results in the disruption of an inhibitory structure that sequesters the Shine-Dalgarno sequence (Fröhlich and Vogel, 2009). We performed in-line probing to determine whether cirA regulation by RyhB and Hfq occurred through a similar mechanism. This technique exploits the natural susceptibility of unpaired and unstructured nucleotides of RNA to spontaneous cleavage in solution (Wakeman and Winkler, 2009). Results showed that RyhB pairing with 5′-end-labelled cirA mRNA in the presence of Hfq (confirmed by the cleavage protection of nucleotides −51 to −42) (Figure 7B, lanes 7 and 8) resulted in increased cleavage of nucleotides −41 to −38 and −34 to −31 (Figure 7B, lanes 7 and 8). Notably, these secondary structure changes are also observed when RyhB is added in the absence of Hfq (compare lanes 5 and 6), suggesting that they do not require Hfq to occur. These RyhB effects on cirA mRNA secondary structure were further confirmed by enzymatic and chemical probing (Supplementary Figure S7). Altogether, these structural changes showed that RyhB pairing with cirA mRNA in the presence of Hfq promoted the partial unfolding of SL3 and that resulted in increased accessibility of region −35 to −26, predicted to be bound by Hfq (Figure 2, Hfq II; Supplementary Figure 6B, C) and to act as a potential translational enhancer (Figure 6D). Based on the bulk of these results, it was predicted that this region was more accessible to 30S subunit upon translation initiation.

The addition of RyhB in the absence of Hfq resulted in cleavage induction of nucleotides +5 to +7 (Figure 7B, lanes 5 and 6), which was previously characterized as a Hfq binding site (Figure 2, Hfq III). Unexpectedly, cleavage of this region was observed when cirA mRNA was incubated with Hfq alone (Figure 7B, lanes 5 and 7). This induced cleavage may result from residual RNase activity in the Hfq preparation (Supplementary Figure S8, lanes 1 and 2). However, it is more probable that the high susceptibility of region +5 to +7 to cleavage upon incubation with RyhB and Hfq (Figure 7B, lane 8) may be in large part attributable to structural changes specific to RyhB pairing, as RyhB alone promoted the same cleavage intensity in this region (Figure 7B, lanes 6 and 8). These data then suggested that RyhB pairing with cirA mRNA in the presence of Hfq resulted in the unfolding of Hfq III region (nucleotides +4 to +6), as it was shown for Hfq II. Since this region is located immediately downstream of the AUG start codon, increased accessibility of this region upon RyhB pairing may stimulate initiation complex formation. Unexpectedly, the structure of the Shine-Dalgarno region (nucleotides −11 to −7) remained unchanged upon RyhB pairing with cirA mRNA in the absence or in the presence of Hfq (Figure 7B, lane 5 and lane 6, lane 7 and lane 8), thereby discarding the possibility that RyhB may activate cirA mRNA translation by making the Shine-Dalgarno more accessible to the ribosome.

Taken together, results described above strongly suggested that RyhB pairing with cirA mRNA resulted in increased accessibility of regions predicted to be targeted by Hfq to repress translation when RyhB was not expressed (Hfq II and Hfq III sites). Indeed, when we mutated the Hfq II site (mutant M2G), we observed a significant decrease in expression of cirA′-′lacZ translational fusion (Supplementary Figure S9). Overall, the data provided a plausible mechanism underlying the increased ribosome binding to cirA mRNA upon RyhB pairing in the presence of Hfq.

RyhB promotes colicin Ia sensitivity through cirA activation

RyhB was expected to modulate colicin sensitivity upon iron starvation by activating cirA translation because colicin Ia is one of the main ligands bound by CirA. WT cells that accumulate high levels of cirA mRNA under conditions of iron starvation should be sensitive to colicin treatment as opposed to ΔryhB cells that are expressing low levels of cirA mRNA. We monitored the growth of WT, ΔryhB and ΔcirA cells in an iron-free minimal medium in the presence or absence of colicin Ia (added at early exponential phase). As expected, WT cells growth was impaired when colicin Ia was added to the medium (Figure 8). In contrast, the growth of ΔryhB and ΔcirA cells was not significantly affected. We explain the observed recovery phenotype of WT+colicin by the active division of a small cell population that was not exposed to colicins at the beginning of the treatment. No suppressor mutations conferring colicin Ia resistance were observed during the course of the experiment (Supplementary Figure S10).

Figure 8.

RyhB expression promotes colicin Ia sensitivity under conditions of iron starvation. Strains EM1055 (WT), EM1238 (ΔryhB), and HS221 (ΔcirA) were grown in M63 iron-free glucose minimal medium and cleared colicin Ia lysate was added at an OD₆₀₀ of 0.1. When needed, 2,3-dihydroxybenzoic acid (DHB, 33 μM) was added 30 min before addition of colicin Ia. For a determination of the levels of RyhB and cirA RNAs performed under the same growth conditions, please refer to Figure 1C.

Because the siderophore DHB is an additional substrate of CirA, we set up a series of experiments to determine whether prebinding of DHB to CirA could prevent the binding of colicin Ia and protect WT cells from the bactericidal effect of colicin Ia. Results demonstrated that the addition of DHB 30 min before colicin Ia treatment of WT cells conferred nearly complete resistance to colicin Ia (Figure 8). Taken together, these data confirmed that RyhB promoted colicin Ia sensitivity under conditions of iron starvation by activating cirA translation.

Discussion

Results reported here unveiled key elements of a complex regulatory network controlling the cellular sensitivity to the antibiotic colicin Ia in an iron-dependent manner (Figure 9). Our data support the interpretation that, under conditions of iron sufficiency, Fe²⁺-Fur (active Fur) repressed cirA and ryhB transcription, which resulted in low levels of CirA. However, under conditions of iron starvation, Fur became inactive and cirA and ryhB were actively transcribed. Strikingly, although cirA mRNA was expressed, Hfq binding to sites II and III prevented formation of the translation initiation complex, leading to rapid cirA destabilization by RNase E. Hfq-induced translational repression was relieved only when sRNA RyhB base paired with cirA mRNA. RyhB allowed translation initiation of cirA mRNA by promoting structural changes that increased the accessibility of ribosomal subunit 30S to sites II (translation enhancer) and III (RBS). Translational activation of cirA mRNA resulted in increased accumulation of CirA protein in the outer membrane, thereby making cells sensitive to the antibiotic colicin Ia.

Figure 9.

Working model for RyhB-mediated translational activation of cirA mRNA (see the text for details). Fe²⁺, ferrous iron; TE, translational enhancer. I, II, and III refer to Hfq binding sites I, II, and III characterized in the study, respectively.

Comparison with other positive sRNA-dependent regulations

One may assume that the regulation of cirA mRNA by RyhB and Hfq is highly similar to other cases of sRNA-mediated translational activation previously characterized (e.g., RyhB-shiA; DsrA-rpoS; ans GlmZ-glmS). In such instances, sRNA pairing with the target mRNA typically results in the disruption of a secondary structure that is detrimental to translation initiation (Fröhlich and Vogel, 2009). However, some key differences highlight each mechanism from the others. First, while Hfq represses cirA and shiA translation when RyhB is not expressed, it does not seem to be the case for rpoS and glmS. Indeed, the hfq mutant strain does not exhibit increased levels of rpoS and glmS translation under conditions of low DsrA and GlmZ expression, respectively (Sledjeski et al, 2001; Urban and Vogel, 2008). Second, RyhB and Hfq are not required for the translation of cirA and shiA. In other words, the secondary structure of both mRNAs is not predicted to prevent translation initiation, according to in vitro and in vivo data. Similar levels of 30S binding to cirA mRNA were observed whether RyhB and Hfq were present or not (Figure 7A). Moreover, cirA and shiA transcripts accumulated to WT levels in Δhfq ΔryhB cells (Figure 5A) and the levels of cirA and shiA translation were identical to WT cells in a Δhfq ΔryhB background (Figure 5B; Prévost et al, 2007). These results were in marked contrast with those of DsrA-rpoS and GlmZ-glmS regulations in which case sRNA and Hfq have been shown to be essential for translation of the target mRNA (Sledjeski et al, 2001; Urban and Vogel, 2008; McCullen et al, 2010). Third, whereas RyhB-shiA, DsrA-rpoS and GlmZ-glmS translational activations occurred through increased accessibility of the Shine-Dalgarno sequence following the sRNA pairing (Prévost et al, 2007), it did not appear to be the case for RyhB-cirA regulation.

Indeed, no structural changes in the Shine-Dalgarno region were observed upon RyhB pairing with cirA mRNA in the presence or in the absence of Hfq (Figure 7B; Supplementary Figure S7). Considering that cirA mRNA can be translated to WT levels in the absence of RyhB and Hfq (Figure 5B), we may assume RBS to be available for ribosome binding upon translation initiation, which is supported by probing data (Supplementary Figure S7B, lane 5; note the high susceptibility of nucleotides G-11, G-10, and G-8 to RNase T1 cleavage) and toeprinting experiments (Figure 7A; Supplementary Figure S6). Remarkably, most structural changes observed upon RyhB pairing to cirA mRNA occurred in regions targeted by Hfq for translational repression (sites II and III, see Figure 2). This result suggested that RyhB action may aim to dislodge Hfq from these sites (discussed below) instead of increasing ribosome access to the Shine-Dalgarno sequence. The incapacity of RyhB to increase 30S binding to cirA mRNA in the absence of Hfq further supported this hypothesis (Supplementary Figure S6).

Antagonistic regulation of cirA translation by Hfq and RyhB

Our work demonstrated a situation in which RyhB activated cirA translation by preventing Hfq binding to a potential translational enhancer element (site II, see Figure 2). To our knowledge, a similar mechanism of sRNA-mediated gene activation has not yet been characterized. The presence of AU-rich sequences located several nucleotides upstream of the Shine-Dalgarno sequence has been reported to enhance translation and to stabilize transcripts (Zhang and Deutscher, 1992; Hook-Barnard et al, 2007). These translational enhancer elements have been shown to increase interaction with ribosomal protein S1 (Boni et al, 1991; Komarova et al, 2002) that typically recognizes AU-rich single-stranded sequences (Ringquist et al, 1995; Hajnsdorf and Boni, 2012). We predict the AU-rich region −35 to −26 of cirA mRNA that we identified as a Hfq binding site (Hfq site II) to act as a translational enhancer, given its close proximity to the Shine-Dalgarno sequence (Figure 2) and the high accessibility of nucleotides −33 to −31, according to PbAc probing (Figure 6C, lane 5). Moreover, mutation of nucleotides U-31 and U-32 to guanines reduced 30S binding by 30% (Figure 6D). This decrease corresponded to the level of repression observed upon incubation of cirA mRNA with Hfq (Figure 7A), suggesting that most of the translational repression mediated by Hfq could occur through binding to these nucleotides. Our findings are in agreement with these observations and suggest a potential role of Hfq II site as a translational enhancer.

Although the 30% repression of 30S binding observed upon incubation of cirA mRNA with Hfq is moderate, we suggest that it is sufficient to repress translation and to promote cirA transcript destabilization. The 60% activation by RyhB observed in toeprint experiments (Figure 7A) would then allow translation to resume and cirA mRNA to accumulate. Further repression of ribosome binding observed with higher concentrations of Hfq (Figure 6A) may not occur in vivo due to limiting concentrations of Hfq in the cell (Moon and Gottesman, 2011; Hussein and Lim, 2011). Moreover, the Hfq binding site III (Figure 2, region +4 to +9) of cirA mRNA may be critical for translational regulation. Given that any sRNA pairing immediately upstream of the Shine-Dalgarno sequence or within the first five codons of the open reading frame was able to compete directly with initiating ribosomes (Bouvier et al, 2008), we would predict Hfq binding to site III to interfere with translation initiation complex formation. Our probing data suggested that Hfq may be dislodged from sites II and III following RyhB pairing (Figure 7B; Supplementary Figure S7). Considering this possibility, we hypothesized that structural changes induced by RyhB upon base pairing with cirA mRNA destabilized Hfq interactions with sites II and III, thereby resulting in translational activation.

It is tempting to suggest that structural changes occurring on cirA mRNA following RyhB pairing not to have major effects on translation in the absence of Hfq, as suggested by toeprint experiments (Supplementary Figure S6). However, RyhB expression in Δhfq cells resulted in a significant increase in cirA mRNA levels (Figure 5C). This Hfq-independent effect of RyhB on cirA expression remains elusive. One may assume this activation to result from the subtle 15% increase in cirA translation observed upon RyhB expression in the absence of Hfq (Figure 5D) that could be sufficient to stabilize cirA mRNA. On the other hand, the possibility of an RyhB-mediated stabilization of cirA transcript independent of ribosome protection cannot be excluded.

Introduction of mutations into the Hfq binding site I did not reduce 30S binding to cirA mRNA (Supplementary Figure S5), suggesting that this region may not regulate translation initiation. Based on the observation that Hfq greatly stimulated RyhB pairing to cirA mRNA in vitro (Supplementary Figure S11), we predict Hfq to bind Hfq I site to promote RyhB-cirA duplex formation in vivo. Upon pairing with cirA mRNA, the RyhB poly(U) tail, which was reported to bind Hfq with high affinity (Otaka et al, 2011), is oriented towards Hfq site I. Binding of Hfq to both site I and RyhB 3′-end under these conditions could then promote stabilization of the RyhB-cirA hybrid, thus allowing sustained translational activation during iron starvation.

A feed-forward loop motif controls colicin sensitivity and iron uptake

The Fur-RyhB-cirA regulatory circuit forms a coherent type 2 feed-forward loop in which Fe²⁺-Fur represses cirA and ryhB while RyhB activates cirA (Supplementary Figure S12). Such a network motif is predicted to delay the induction of the target gene, according to computational modelling on transcription networks (Mangan and Alon, 2003). Application of the model to cirA regulation when cells experience iron starvation predicts cirA expression to be delayed in comparison to direct regulation by Fe²⁺-Fur. This would be logical, considering that once Fur becomes inactive and cirA promoter is fully expressed, RyhB synthesis must reach a certain threshold for cirA mRNA to be translated and stabilized. Conversely, no lag of repression is expected for transcriptional networks with the same configuration as the Fur-RyhB-cirA loop. This prediction may not apply to cirA repression, considering the fact that feed-forward loops integrating sRNAs often display altered regulatory dynamics when compared to transcriptional feed-forward loops (Shimoni et al, 2007; Beisel and Storz, 2010, 2011). One could imagine that under conditions when cells switch from iron-poor to iron-rich conditions and when Fur becomes active, despite the fact that Fe²⁺-Fur represses cirA and ryhB transcription, there are still cirA transcripts activated by existing RyhB molecules that were transcribed before Fur repression. Consequently, CirA proteins will be synthesized until the existing pool of cirA transcripts becomes exhausted. We would then expect cirA repression to be delayed in this context as compared to direct repression by Fe²⁺-Fur.

Physiological relevance of colicin sensitivity upon iron starvation

Colicins are produced following DNA damages as a result of the SOS response as well as various cellular stresses such as thymine starvation, high temperatures or anaerobiosis (Cascales et al, 2007). Under these conditions, colicin-producing strains lyse and released colicins kill surrounding colicin-sensitive cells. Despite the fact that the role of colicins in survival to iron starvation has not yet been addressed, we can speculate colicin sensitivity to confer a selective advantage for the bacterial community when iron is scarce. Cells experiencing iron deficiency and, thereby expressing CirA, would become sensitive to the killing action of colicin Ia and lyse. Then, the released iron could be imported from the environment by the remaining cell population for survival. The DHB molecules expelled in the medium upon cell death could in turn protect the remaining cells from colicin action through competitive binding to CirA, as observed here (Figure 8 in the case of WT cells), thus preventing elimination of the whole bacterial population.

Our work further supports the active role of RyhB in gene activation in addition to its well-known function as a gene silencer. The feed-forward loop by the Fur-RyhB-cirA regulatory circuit also unveils a novel type of iron-dependent regulation. To our knowledge, cirA is the first Fur-regulated gene to require further post-transcriptional activation by RyhB upon iron starvation that has been characterized so far. Further investigation is warranted to uncover similar type of regulation in the future.

Materials and methods

Strains and plasmids

Derivatives of E. coli MG1655 were used in all experiments (Supplementary Table S1). DH5α strain was used for routine cloning procedures. EM1055 (wild type, MG1655 derivative), EM1451 (Δara714 leu⁺), EM1455 (Δara714 leu⁺ ΔryhB::cat), EM1238 (ΔryhB::cat), EM1256 (Δfur::kan), EM1237 (DY330 [W3110 ΔlacU169 gal490 λcI857 Δ(cro-bioA)]), EM1265 [hfq-1::Ω(kan;Bcl1)], EM1277 (rne-3071 zce-726_Tn10), EM1280 (rne-3071 zce-726_Tn10 ΔryhB::cat), EM1377 (rne-131 zce-726_Tn10), KP392 (Δfur::kan), KP393 (Δfur::kan ΔryhB::cat), and JW4130-1 (rrnB3 ΔlacZ4787 hsdR514 Δ(araBAD)567 Δ(rhaBAD)568 rph-1 Δhfq-722::kan) have been described earlier (Yu et al, 2000; Massé and Gottesman, 2002; Massé et al, 2003; Baba et al, 2006; Desnoyers et al, 2009; Salvail et al, 2010). Strains constructed by P1 transduction were selected for the appropriate antibiotic-resistant marker. Except as otherwise indicated, for cells carrying plasmid pNM12, pBAD-ryhB, pBAD-ryhB1 and pCP20 (see Supplementary Table S2 for a list of plasmids), ampicillin was used at a final concentration of 50 μg/ml. Deletion/insertion mutations in cirA (ΔcirA::tet), omrAB (ΔomrAB::tet), araB (ΔaraB::kan) and fur (Δfur::tet) were constructed by the method described by Yu et al. Tetracycline resistance cassette was first amplified by PCR from strain EM1053 (oligos EM215-EM216) (see Supplementary Table S3 for a list of oligonucleotides). The PCR fragment obtained was then amplified with oligos containing sequences that are homologous to the 5′ and 3′ ends of the cassette and sequences homologous to cirA (oligos EM1044–EM1045), omrAB (oligos EM1599–EM1600), and fur (oligos EM1989–EM1990). Flippase recognition target (FRT)-flanked kanamycin resistance cassette was generated by PCR from a DH5α strain harbouring pKD4 plasmid with oligos containing sequences that are homologous to the 5′ and 3′ ends of the cassette and sequences homologous to araB (oligos EM1388–EM1389). The resulting PCR products were transformed into EM1237 after induction of λred, according to Yu et al, selecting for tetracycline resistance for cirA and omrAB mutations and for kanamycin resistance for araB mutation. Recombinant products were verified by sequencing. Following P1 transduction of araB::kan, kanamycin resistance cassette was removed using flippase (Flp) encoding helper plasmid pCP20 as described (Datsenko and Wanner, 2000).

RNA extraction and northern blot analysis

Northern blots were performed as described previously (Desnoyers and Massé, 2012) with some modifications. Total RNA was extracted from cells grown at 37°C or 30°C, for experiments with thermosensitive strains (EM1277 and EM1280), in LB or M63 minimal medium containing glucose or glycerol (0.2%) using the hot phenol procedure (Aiba et al, 1981). Arabinose (0.1%), 2,2′-dipyridyl (250 μM) or FeSO₄ (5 μM) was added when indicated. Following total RNA extraction, 5–10 μg of total RNA was loaded on polyacrylamide gel (4–6% acrylamide 29:1, 8 M urea) and 15 μg was loaded on agarose gel (1%, 1X MOPS).

Primer extension

Reverse transcription was performed as previously described (Prévost et al, 2007).

OMPs analysis

Cells were grown in 50 ml of LB medium to an OD₆₀₀ of 1.0 and 10 ml of culture was harvested and OMPs were extracted as previously described (Morona and Reeves, 1982; Guillier and Gottesman, 2006).

β-Galactosidase assays

Kinetic assays for β-galactosidase activity were performed as described (Prévost et al, 2007), using a SpectraMax 250 microtitre plate reader (Molecular Devices).

qRT–PCR analysis

qRT–PCR was performed according to a previous report (Salvail et al, 2010).

RNA secondary structure probing

Secondary structure probing was performed as described earlier (Desnoyers et al, 2009). In-line probing experiments were performed as previously described (Regulski and Breaker, 2008; Wakeman and Winkler, 2009).

Toeprinting assays

Toeprinting assays were carried out according to a previous report (Salvail et al, 2010).